Riemann Hypothesis Proof Claimed: AI-Human Breakthrough

Isometric editorial illustration of a dark sphere resting on a long cream bar with a thin oxblood line, evoking a machine-assisted mathematical proof.
Image via Wisevoter

Key takeaways

  • Mathematician Levent Alpöge claims to have discovered a Riemann Hypothesis proof using Anthropic’s Claude AI.
  • The discovery includes a solution to the Generalized Riemann Hypothesis (GRH).
  • The proof is presented via 14 archive files containing over 175,000 Lean 4 formal verification files.
  • The mathematical community has not yet independently verified the validity of the claim or the logic.

On November 20, 2026, mathematician Levent Alpöge announced a potential Riemann Hypothesis proof that was developed through a collaborative effort with Anthropic’s Claude AI model. The claim, which was published on Alpöge’s personal website, asserts that the discovery resolves not only the classical Riemann Hypothesis but also the more expansive Generalized Riemann Hypothesis (GRH). While the announcement has sent ripples through the mathematical community, experts emphasize that the claim remains unverified, as no formal peer-reviewed paper has been released and the provided Lean code has yet to be independently audited for logical soundness.

What happened

Levent Alpöge, a mathematician at the AI safety company Anthropic and a former Harvard fellow, officially posted his findings on Friday, November 20, 2026. Rather than releasing a traditional, long-form prose manuscript, Alpöge provided a digital archive consisting of 14 compressed files. This archive, totaling approximately 239 megabytes, contains more than 175,000 individual source files written in Lean 4, a formal programming language specifically designed for machine-verifiable mathematics.

According to Alpöge, the process was a hybrid of human-directed inquiry and high-level AI execution. He described a scenario where he provided the initial conceptual direction, suggesting a mathematical setting based on two specific ideas from a simpler model. However, once the direction was established, the Claude AI model performed the heavy technical lifting. Alpöge noted that during the final stages of the process, the AI’s ability to navigate complex mathematical structures, specifically in the realm of automorphic forms on higher rank groups, far outpaced his own immediate mastery of the subject.

In a striking departure from traditional academic norms, Alpöge opted not to publish a human-readable written argument. He explained that an informal, AI-generated version of the proof had reached a staggering length of over 250,000 pages, making it an “unreadable travesty” for human consumption. Instead, he has provided the formal logic, trusting the Lean kernel to serve as the ultimate arbiter of truth. This move highlights a fundamental shift in how scientific breakthroughs may be documented and verified in an age of AI-driven research.

Close-up of colorful source code on a monitor, showcasing programming and technology
Photo by Abdul Kayum on Pexels

Why it matters

The Riemann Hypothesis has remained one of the most significant unsolved problems in mathematics for 167 years. First proposed by Bernhard Riemann in 1859, the hypothesis concerns the distribution of prime numbers. It posits that all non-trivial zeros of the Riemann zeta function lie on a specific vertical “critical line” in the complex plane. Because the location of these zeros dictates the error margin in how prime numbers are distributed, a proof would provide a definitive foundation for much of modern number theory.

If the Riemann Hypothesis proof is verified, the impact on the mathematical landscape would be profound. For over a century, mathematicians have built vast architectures of theorems that are currently “conditional,” meaning they are only assumed to be true if the Riemann Hypothesis is true. A successful proof would turn these conditional statements into unconditional mathematical certainties. This would provide absolute clarity regarding the predictable patterns hidden within the seemingly random distribution of prime numbers.

Furthermore, the discovery carries significant financial and professional stakes. The Clay Mathematics Institute has designated the Riemann Hypothesis as one of the seven Millennium Prize Problems, offering a $1 million reward for its solution. However, the Institute requires that a solution be published in a qualifying journal and undergo a two-year period of general acceptance by the mathematical community before the prize is awarded.

The role of Claude in the discovery

The methodology employed by Alpöge represents a new paradigm: human-directed, AI-assisted research. This model differs from previous uses of AI, which typically focused on data processing or simple pattern recognition. In this instance, the AI was a collaborator in high-level theoretical reasoning.

Alpöge credited Claude with making rapid technical advances in areas where he admitted to having “little mastery.” Specifically, the AI was instrumental in constructing a complex mathematical mechanism involving a “von Mangoldt sum” and a long chain of “theta lifts.” The length of this chain, according to the claim, scales according to the square root of the size of the primes being counted. This specific technical construction was the key to resolving the issue of “multiplicity,” a major hurdle that has prevented previous attempts to bridge the gap between simpler models and the full Riemann Hypothesis.

This level of autonomy in an AI model is unprecedented. Alpöge described the experience of watching the AI’s progress as one that left him “in awe.” It suggests a future where researchers act as architects or directors, setting the conceptual boundaries and objectives, while AI agents act as the specialized engineers that execute the intricate, high-volume technical work required to reach a conclusion.

Technical deep dive: Lean 4 and formal verification

To understand why Alpöge released code instead of a paper, one must understand the role of Lean formal verification. In traditional mathematics, a proof is a series of logical arguments written in natural language. These proofs are checked by human peers, a process that is prone to error, especially as proofs become increasingly long and complex.

Lean 4 is a “proof assistant” that uses a formal language to ensure logical soundness. When a proof is written in Lean, every single step must be checked by a “Lean kernel,” a small, highly trusted piece of software that verifies the logical validity of every deduction. This eliminates the possibility of human oversight or subtle logical leaps that might go unnoticed in a traditional manuscript.

However, the mathematical community has raised several critical points regarding the use of Lean in this context. Experts emphasize that a successful “build” in Lean does not automatically mean the hypothesis has been proven. There are four essential verification steps that independent researchers must undertake:

  1. Theorem Statement Audit: Researchers must ensure that the definitions used in the code—such as the zeta function and the critical line—are the standard mathematical definitions and have not been subtly altered to make the proof easier to achieve.
  2. Search for “Sorry” or Custom Axioms: In Lean, the command sorry is used as a placeholder for a step that has not yet been proven. A valid proof cannot contain these. Additionally, researchers must check if the proof relies on custom axioms that bypass necessary logical steps.
  3. Clean Build Test: The code must be able to compile on a fresh, independent machine to ensure it does not rely on undocumented local files or specific environments used by the author.
  4. Dependency Inspection: The proof must be checked to ensure it relies on standard, verified mathematical libraries rather than unreviewed, local files that might contain errors.
Three adults solving complex mathematical equations on a chalkboard indoors.
Photo by Jean-Rene Chazottes on Pexels

Simultaneous claims: The Navier–Stokes breakthrough

In a remarkable coincidence, Alpöge’s announcement regarding the Riemann Hypothesis was accompanied by a second major claim involving fluid dynamics. Alongside co-author Tristan Buckmaster, Alpöge released a 79-page paper claiming to have solved the Navier–Stokes existence and smoothness problem.

This problem, another Millennium Prize challenge, asks whether solutions to the Navier–Stokes equations—which describe the motion of fluid substances—always remain smooth and well-behaved, or if they can develop “singularities” (points where the fluid’s velocity or pressure becomes infinite).

Feature Riemann Hypothesis Claim Navier–Stokes Claim
Primary Goal Prove zeros of zeta function lie on critical line Prove fluid singularities develop in unforced equations
Key Tool Claude AI & Lean 4 Code 79-page paper & numerical certificates
Status Unverified (Code pending audit) Unverified (Certificate pending release)
Collaborators Levent Alpöge & Claude Levent Alpöge & Tristan Buckmaster
Mathematical Field Number Theory Fluid Dynamics

Unlike previous claims in this field, such as those made by OpenAI in September 2026 regarding “forced” Navier–Stokes equations, Alpöge and Buckmaster claim that the breakdown occurs under the fluid’s own evolution, without any external force being applied. This distinction is vital, as a proof of singularity in the unforced case would fundamentally alter the scientific understanding of fluid behavior in three-dimensional space.

Controversy: Timing and the Anthropic IPO

The timing of the announcement has not escaped scrutiny. Alpöge’s claim arrived just as Anthropic’s shares began trading on the Nasdaq, following an IPO priced at $83. This has led to speculation on platforms like Hacker News that the announcement might have been timed to influence market perception of Anthropic’s technological capabilities.

Alpöge has addressed these allegations directly, calling the suggestion that the timing was intentional “cynical” and “false.” He explained that the results were finalized late on a Sunday night and that he had to navigate certain constraints regarding how and when he could announce the findings. He maintained that his decision to proceed was shared with Anthropic leadership and was driven by the mathematical significance of the discovery rather than financial considerations.

What it means for the scientific community

The implications of this event extend far beyond the field of number theory. If verified, this represents a monumental shift in the methodology of scientific research. We are entering an era where the “human-directed, AI-assisted” model could become the standard for tackling the world’s most complex problems.

For mathematicians, this means a new set of skills will be required. The ability to write and audit formal code in languages like Lean may become as essential as the ability to write traditional proofs. For the broader scientific community, it signals that AI is moving from a tool of assistance to a tool of primary discovery, capable of navigating mathematical “neighborhoods” that are too vast or complex for a single human mind to master quickly.

However, this shift also introduces the “black box” problem. If an AI produces a proof that is too long or complex for any human to read or intuitively grasp, science faces a crisis of understanding. We may reach a point where we can verify that a statement is true through machine code, but we may no longer understand why it is true. This tension between formal truth and human comprehension will likely be a central theme in the coming decade of scientific inquiry.

Counterpoints and open questions

Despite the excitement, significant skepticism remains. The primary concern is the potential for encoding errors. As noted by technical observers at OfficeChai, a Lean proof is only as valid as the theorem statement it is proving. If the code accurately proves a modified or simplified version of the Riemann Hypothesis rather than the actual conjecture, the result would be mathematically meaningless.

There are also concerns regarding transparency. For the Navier–Stokes claim, the public repository currently omits the critical “numerical certificate” and the interval-arithmetic computations required for reproduction. Without these, other researchers cannot independently verify the bounds of the proof.

Furthermore, the sheer scale of the AI-generated output raises questions about the nature of credit. As noted by observers on Hacker News, the division of labor between human and AI is difficult to quantify. If Claude is responsible for the bulk of the technical execution, how should the mathematical community attribute authorship? This ambiguity challenges long-standing traditions of academic recognition and intellectual property.

A magnifying glass and pencil on a hand-drawn architectural sketch pad, suggesting
Photo by Tima Miroshnichenko on Pexels

What happens next

The coming months will be a period of intense scrutiny for Alpöge and the mathematical community. The following signals will be critical to watch:

  • Independent Lean Audits: The first major milestone will be whether independent teams can successfully run the 175,000 Lean files and confirm they compile and prove the intended theorem statement.
  • Peer-Reviewed Publication: For the claims to gain formal legitimacy, they must be published in recognized mathematical journals, a process that can take months or even years.
  • Clay Mathematics Institute Review: Should the proofs pass the peer-review stage, the Institute will begin its rigorous two-year evaluation process to determine if the $1 million Millennium Prize should be awarded.
  • Expansion to ERH: Mathematicians will look to see if Alpöge can successfully extend this methodology to the Extended Riemann Hypothesis (ERH), as he has indicated is his intent.

As the mathematical community begins the arduous task of auditing these massive, machine-checked files, the world will wait to see if this marks the resolution of a 167-year-old mystery or serves as a cautionary tale about the complexities of AI-driven science. Whether the claim holds or fails, the event has already fundamentally altered the conversation regarding the future of human and machine intelligence in the pursuit of truth.

Frequently asked questions

What is the Riemann Hypothesis?

The Riemann Hypothesis is a mathematical conjecture proposed by Bernhard Riemann in 1859 regarding the distribution of prime numbers. It suggests that all non-trivial zeros of the Riemann zeta function lie on a specific vertical line in the complex plane, known as the “critical line.” If true, it would confirm that prime numbers follow a highly structured pattern rather than being distributed randomly.

How did Claude AI assist in the proof?

Levent Alpöge used Claude as a collaborative partner to handle the high-level technical execution of the proof. While Alpöge provided the conceptual framework and direction, the AI was responsible for navigating complex mathematical areas, such as automorphic forms, and constructing the technical chains of logic required to complete the argument.

Is the Riemann Hypothesis proof verified?

No, the proof is currently unverified. While Alpöge has released the formal logic in the Lean 4 programming language, the broader mathematical community has not yet conducted an independent audit to ensure that the theorem was encoded correctly and that the logic within the code is sound.

What is Lean 4?

Lean 4 is a formal programming language and proof assistant used to express mathematical arguments in a way that can be checked by a computer. It uses a “kernel” to verify every logical step of a proof, providing a level of automated certainty that traditional, human-read mathematical papers cannot offer.

References

Featured image: Image via Wisevoter

One thought on “Riemann Hypothesis Proof Claimed: AI-Human Breakthrough”

Leave a Reply